def next_batch(bs=batch_size, train=True):
x_batch = []
y_batch = []
seq_len_batch = []
original_batch = []
i = 0
for k in range(bs):
ut_length_dict = dict([(k, len(v['target']))
for (k, v) in audio.cache.items()])
utterances = sorted(ut_length_dict.items(), key=operator.itemgetter(0))
test_index = 346
if train:
utterances = [a[0] for a in utterances[test_index:]]
else:
utterances = [a[0] for a in utterances[:test_index]]
training_element = audio.cache[utterances[i]]
target_text = training_element['target']
audio_buffer = training_element['audio']
x, y, seq_len, original = convert_inputs_to_ctc_format(
audio_buffer, sample_rate, 'whatever', num_features)
x_batch.append(x)
y_batch.append(y)
seq_len_batch.append(seq_len)
original_batch.append(original)
i += 1
y_batch = sparse_tuple_from(y_batch)
seq_len_batch = np.array(seq_len_batch)[:, 0]
for i, pad in enumerate(np.max(seq_len_batch) - seq_len_batch):
x_batch[i] = np.pad(x_batch[i], ((0, 0), (0, pad), (0, 0)),
mode='constant',
constant_values=0)
x_batch = np.concatenate(x_batch, axis=0)
return x_batch, y_batch, seq_len_batch, original_batch
def next_batch(bs=batch_size, train=False):
x_batch = []
y_batch = []
seq_len_batch = []
original_batch = []
ut_length_dict = dict([(k, len(v['target']))
for (k, v) in audio.cache.items()])
utterances = sorted(ut_length_dict.items(), key=operator.itemgetter(0))
utterances = [a[0] for a in utterances[:]]
for k in range(bs):
training_element = audio.cache[utterances[k]]
target_text = training_element['target']
audio_buffer = training_element['audio']
x, y, seq_len, original = convert_inputs_to_ctc_format(
audio_buffer, sample_rate, target_text, num_features)
x_batch.append(x)
y_batch.append(y)
seq_len_batch.append(seq_len)
original_batch.append(original)
# Creating sparse representation to feed the placeholder
# inputs = np.concatenate(x_batch, axis=0)
y_batch = sparse_tuple_from(y_batch)
seq_len_batch = np.array(seq_len_batch)[:, 0]
for i, pad in enumerate(np.max(seq_len_batch) - seq_len_batch):
x_batch[i] = np.pad(x_batch[i], ((0, 0), (0, pad), (0, 0)),
mode='constant',
constant_values=0)
x_batch = np.concatenate(x_batch, axis=0)
# return np.array(list(x_batch[0]) * batch_size), y_batch, np.array(seq_len_batch[0] * batch_size), original_batch
# np.pad(x_batch[0], ((0, 0), (10, 0), (0, 0)), mode='constant', constant_values=0)
return x_batch, y_batch, seq_len_batch, original_batch
def next_testing_file(file_name):
from audio_reader import read_audio_from_filename
truncated_audio = read_audio_from_filename(
file_name, sample_rate=c.AUDIO.SAMPLE_RATE)
from utils import convert_inputs_to_ctc_format
test_inputs, targets, test_seq_len, original = convert_inputs_to_ctc_format(
truncated_audio, c.AUDIO.SAMPLE_RATE, "")
return test_inputs, test_seq_len, original
def next_training_batch():
import random
from utils import convert_inputs_to_ctc_format
random_index = random.choice(list(audio.cache.keys())[0:5])
training_element = audio.cache[random_index]
target_text = training_element['target']
out = convert_inputs_to_ctc_format(training_element['audio'],
c.AUDIO.SAMPLE_RATE, target_text)
train_inputs, train_targets, train_seq_len, original = out
return train_inputs, train_targets, train_seq_len, original
def next_testing_batch():
import random
from utils import convert_inputs_to_ctc_format
random_index = random.choice(list(audio.cache.keys())[0:5])
training_element = audio.cache[random_index]
target_text = training_element['target']
random_shift = np.random.randint(low=1, high=1000)
print('random_shift =', random_shift)
truncated_audio = training_element['audio'][random_shift:]
train_inputs, train_targets, train_seq_len, original = convert_inputs_to_ctc_format(
truncated_audio, c.AUDIO.SAMPLE_RATE, target_text)
return train_inputs, train_targets, train_seq_len, original, random_shift
